The paperwork shortcut has a paperwork problem. AI transcriptions are no time-saving measure for doctors when the resulting notes need checking, correcting, and comparing with what patients actually said.
Dr Mary Gibbs, an out-of-hours GP, says AI has a tendency to misunderstand and contradict itself, creating dangers for the patient. Many patients seen by doctors have already been assessed by someone else on the phone, so the later consultation depends on a history that must be clear, accurate, and complete.
AI consultations are invariably long, filled with duplications and sometimes contradictions. Any history taken by AI can turn out very different from what is listed, while meaningful errors appear more often in AI consultations than in notes typed by colleagues.
That is not a minor editing nuisance. Turning a complex patient history into a clear account is a difficult skill to learn, and the process involves more than capturing words. A note must preserve the details that help medical staff understand what happened and decide what matters.
Accuracy Is Not the Only Problem
Debbie Cameron, a secretary, experienced the limits of voice recognition software firsthand. The system misinterpreted “two weeks on lansoprazole” as “a two-week holiday on Lanzarote” — an amusing error until the subject is a patient’s treatment rather than someone’s travel plans.
AI tools used to take medical notes from patients can miss vital information and lead to medical staff being deskilled. AI scribes are used by about 40% of GPs in the UK, or up to four in 10, giving these weaknesses a large place in routine consultations.
The missing information may not be spoken words. AI tools can miss facial expressions, gestures, and emotional state, all of which can form part of a consultation. Patients may also be less forthcoming if they know they are being recorded and processed by AI, changing the conversation before the system produces its summary.
That creates a problem for documentation as well as trust. AI summaries tend to prioritize clinical information at the expense of patients’ stories and experiences, raising the risk that patients’ stories are lost and their voices are not preserved in AI documentation.
The Cost of Offloading Clinical Judgment
Dr Lucas Seuren, a research fellow at the University of Edinburgh’s Centre for Biomedicine, Self and Society, says, “Many clinicians are excited about ambient AI scribes, because they promise to cut down on paperwork.” The promise is clear; the trade-off is easier to miss.
Using AI scribes could create “cognitive offloading” for clinicians, reducing memory recall and skill development. Manual notetaking serves important functions in reasoning and reflection, so removing that work may remove part of the thinking that happens during a consultation.
The concern is not that every AI-generated note fails. The concern is that the mistakes, omissions, and contradictions can sit inside a document that looks complete, while the manual process that once helped clinicians notice and interpret details becomes weaker.
Seuren puts the patient impact plainly: “The experiences of patients are poorly considered, and there are real risks that the patients’ stories are lost.” A transcription system can produce more words without producing a better account.
For doctors, the issue is not whether AI can recognize speech. It is whether the resulting record preserves vital information, patient experience, and the clinical reasoning needed to use that record safely. So far, the evidence and expert concerns point to a tool that adds work where it promised to remove it.
Based on
- AI transcriptions are no time-saving measure for doctors | Letters — theguardian.com
- Doctors using AI to take patients’ notes failing to capture ‘potentially vital information’ | The Independent — independent.co.uk
- AI tools that take notes for GPs could miss vital information, study suggests — bbc.com
